Application of artificial intelligence in a real-world research for predicting the risk of liver metastasis in T1 colorectal cancer

Abstract Background Liver is the most common metastatic site of colorectal cancer (CRC) and liver metastasis (LM) determines subsequent treatment as well as prognosis of patients, especially in T1 patients. T1 CRC patients with LM are recommended to adopt surgery and systematic treatments rather than endoscopic therapy alone. Nevertheless, there is still no effective model to predict the risk of LM in T1 CRC patients. Hence, we aim to construct an accurate predictive model and an easy-to-use tool clinically. Methods We integrated two independent CRC cohorts from Surveillance Epidemiology and End Results database (SEER, training dataset) and Xijing hospital (testing dataset). Artificial intelligence (AI) and machine learning (ML) methods were adopted to establish the predictive model. Results A total of 16,785 and 326 T1 CRC patients from SEER database and Xijing hospital were incorporated respectively into the study. Every single ML model demonstrated great predictive capability, with an area under the curve (AUC) close to 0.95 and a stacking bagging model displaying the best performance (AUC = 0.9631). Expectedly, the stacking model exhibited a favorable discriminative ability and precisely screened out all eight LM cases from 326 T1 patients in the outer validation cohort. In the subgroup analysis, the stacking model also demonstrated a splendid predictive ability for patients with tumor size ranging from one to50mm (AUC = 0.956). Conclusion We successfully established an innovative and convenient AI model for predicting LM in T1 CRC patients, which was further verified in the external dataset. Ultimately, we designed a novel and easy-to-use decision tree, which only incorporated four fundamental parameters and could be successfully applied in clinical practice.

Download Full-text

Application of Artificial Intelligence in a Real-World Research for Predicting the Risk of Liver Metastasis in T1 Colorectal Cancer

10.21203/rs.3.rs-746689/v1 ◽

2021 ◽

Author(s):

Tenghui Han ◽

Jun Zhu ◽

Dong Xu ◽

Rujie Chen ◽

Shuai Wang ◽

...

Keyword(s):

Artificial Intelligence ◽

Colorectal Cancer ◽

Liver Metastasis ◽

Predictive Model ◽

Tumor Size ◽

Area Under The Curve ◽

Subsequent Treatment ◽

Tumor Type ◽

Discriminative Ability ◽

N Stage

Abstract Background: The liver is the most common metastatic site of colorectal cancer (CRC) and liver metastasis (LM) determines subsequent treatment as well as prognosis of patients, especially in T1 patients. T1 CRC patients with LM are recommended to adopt surgery and systematic treatments rather than endoscopic therapy alone. However, there is still no effective model to predict the risk of LM in T1 CRC patients and we aim to develop a novel and accurate predictive model.Methods: We integrated two independent CRC cohorts from Surveillance Epidemiology and End Results database (SEER) and Xijing hospital. Artificial intelligence (AI) and machine learning methods were adopted to establish the predictive model.Results: A total of 16785 and 326 T1 CRC patients from SEER database and our hospital were incorporated respectively in the study. We found that age, gender, married status, primary site, tumor size, carcinoembryonic antigen (CEA), tumor type, grade, N stage and perineural invasion were significant independent factors for predicting the presence of LM, among which tumor size is the most important. The stacking bagging model showed the best predictive capability, achieving a sensitivity of 0.8452, a specificity of 0.9566, and an area under the curve of 0.9631. In addition, the stacking model had an excellent discriminative ability and accurately screened out eight LM cases from 326 T1 patients in the outer validation cohort. Ultimately, we authenticated the prognostic value of the stacking model, which is consistent with the predictive result of LM.Conclusion: We successfully established an innovative and convenient AI model for predicting LM in T1 CRC patients, which was further verified in our dataset.

Download Full-text

DNA methylation signatures predicting liver metastasis of colorectal cancer: A proof-of-concept pilot study.

Journal of Clinical Oncology ◽

10.1200/jco.2020.38.15_suppl.e16080 ◽

2020 ◽

Vol 38 (15_suppl) ◽

pp. e16080-e16080

Author(s):

Jianming Ying ◽

Weihua Li ◽

Kaihua Liu ◽

Cong Xiao ◽

Shuyu Wu ◽

...

Keyword(s):

Colorectal Cancer ◽

Liver Metastasis ◽

Early Detection ◽

Predictive Model ◽

Primary Tumor ◽

Metastatic Tumor ◽

Training Data ◽

Training Dataset ◽

Tissue Samples ◽

Normal Tissues

e16080 Background: Liver metastasis (LIM) is the leading cause of death in colorectal cancer (CRC) patients. Early detection of LIM may improve outcome in CRC patients. The aim of this study was to evaluate the feasibility of predicting LIM of CRC using methylation profiles. Methods: We performed Roche targeted (~5.5 million methylation sites) bisulfite sequencing of matched primary, metastatic and their adjacent normal tissue samples from 5 CRC patients with LIM, 5 patients with lung metastasis (LUM) and 8 patients without metastasis in the training cohort (n = 48 samples). Differential methylation regions (DMR) of LUM were identified and a predictive model was developed. The model was further validated in primary tumor sample from nine patients (6 with LIM). Results: By comparing primary tumor vs adjacent normal tissues and metastatic tumor vs adjacent normal tissues in CRC patients with LIM, we identified 28954 common DMRs which indicating the methylation characteristic of CRC with LIM. Similarly, 16187 DMRs were identified in patients with LUM. 9179 DMRs are shared in both LIM and LUM comparisons which should be the common characteristic of CRC tumor tissue regardless of the location of metastasis. 7008 DMRs are LUM specific and 19775 DMRs are LIM specific. In order to predict LIM in primary, early changes in LIM specific DMRs should be identified. Hence, we further selected 4134 DMRs by chossing significantly differentically methylated regions between LIM primary tissues and LUM primary tissues. To increase the ability of distinguishing LIM from other normal tissues and non-matastasis CRC tumors, 1215 DMRs were finally selected which also showed increasing or decreasing trend of methylation level through the progression of CRC. The final 1215 biomarkers were used to construct a random forest model using methlylation profile of 5 CRC patients with LIM as positive training data and 5 CRC patients with LUM as well as 8 patients without metastasis as negative training data. Through the feature recursive elimination method, one methylation site (chr8.72468901-72469000) was identified with ROC of 0.9 in the training dataset. The predictive model was validated in an independent dataset which is composed of 6 patients with LIM and 3 patients without metastasis, and achieved an AUC of 0.87. Conclusions: Our findings demonstrate the utility of methylation biomarkers for the molecular characterization of metastatic precursors, with implications for prediction and early detection of liver metastasis in CRC.

Download Full-text

Accurate recognition of colorectal cancer with semi-supervised deep learning on pathological images

Nature Communications ◽

10.1038/s41467-021-26643-8 ◽

2021 ◽

Vol 12 (1) ◽

Author(s):

Gang Yu ◽

Kai Sun ◽

Chao Xu ◽

Xing-Hua Shi ◽

Chong Wu ◽

...

Keyword(s):

Artificial Intelligence ◽

Colorectal Cancer ◽

Deep Learning ◽

Supervised Learning ◽

Area Under The Curve ◽

Patient Level ◽

Significant Difference ◽

The Mean ◽

Whole Slide Images ◽

Better Than

AbstractMachine-assisted pathological recognition has been focused on supervised learning (SL) that suffers from a significant annotation bottleneck. We propose a semi-supervised learning (SSL) method based on the mean teacher architecture using 13,111 whole slide images of colorectal cancer from 8803 subjects from 13 independent centers. SSL (~3150 labeled, ~40,950 unlabeled; ~6300 labeled, ~37,800 unlabeled patches) performs significantly better than the SL. No significant difference is found between SSL (~6300 labeled, ~37,800 unlabeled) and SL (~44,100 labeled) at patch-level diagnoses (area under the curve (AUC): 0.980 ± 0.014 vs. 0.987 ± 0.008, P value = 0.134) and patient-level diagnoses (AUC: 0.974 ± 0.013 vs. 0.980 ± 0.010, P value = 0.117), which is close to human pathologists (average AUC: 0.969). The evaluation on 15,000 lung and 294,912 lymph node images also confirm SSL can achieve similar performance as that of SL with massive annotations. SSL dramatically reduces the annotations, which has great potential to effectively build expert-level pathological artificial intelligence platforms in practice.

Download Full-text

Three-Dimensional Length from the Center of the Liver is a Prognostic Factor of Colorectal Cancer with Liver Metastasis: A Retrospective Analysis

10.21203/rs.3.rs-31845/v1 ◽

2020 ◽

Author(s):

Jun Woo Bong ◽

Yeonuk Ju ◽

Jihyun Seo ◽

Sang Hee Kang ◽

Pyoung-Jae Park ◽

...

Keyword(s):

Colorectal Cancer ◽

Risk Factors ◽

Overall Survival ◽

Liver Metastasis ◽

Neoadjuvant Chemotherapy ◽

Resection Margin ◽

Characteristic Curve ◽

Significant Risk ◽

Area Under The Curve ◽

Positive Resection Margin

Abstract Background Resectability of liver metastasis is important to establish a treatment strategy for colorectal cancer patients. We aimed to evaluate the effect of distance from metastasis to the center of the liver on the resectability and patient outcomes after hepatectomy. Methods Clinical data of a total of 124 patients who underwent hepatectomy for colorectal cancer with liver metastasis were retrospectively reviewed. We measured the minimal length from metastasis to the bifurcation of the portal vein at the primary branch of the Glissonean tree and defined it as “Centrality”. Predictive effects on positive resection margin and overall survival of centrality were statistically analyzed. Results The value as a predictive factor for the positive resection margin of centrality was analyzed by the receiver operating characteristic curve (area under the curve = 0.72, P<0.001). In multivariate analysis, total number of metastases ≥ 3 and centrality ≤ 1.5 cm were significant risk factors of overall survival. Patients with these two risk factors (n=21) had worse 5-year overall survival (10.7%) than patients with one (n=35, 58.3%) or no risk factor (n=68, 69.2%). In subgroups analysis, neoadjuvant chemotherapy improved overall survival in patients with these two risk factors. Conclusion Centrality was related with a positive resection margin and had a negative effect on survival. By combining the total number of metastases with centrality, we could determine disease prognosis and neoadjuvant chemotherapy indications for advanced colorectal cancer with liver metastasis.

Download Full-text

Targeted UPLC-MS Metabolic Analysis of Human Faeces Reveals Novel Low-Invasive Candidate Markers for Colorectal Cancer

Cancers ◽

10.3390/cancers10090300 ◽

2018 ◽

Vol 10 (9) ◽

pp. 300 ◽

Cited By ~ 3

Author(s):

Joaquin Cubiella ◽

Marc Clos-Garcia ◽

Cristina Alonso ◽

Ibon Martinez-Arranz ◽

Miriam Perez-Cormenzana ◽

...

Keyword(s):

Colorectal Cancer ◽

Predictive Model ◽

Precancerous Lesions ◽

Tumour Progression ◽

Haemoglobin Concentration ◽

Area Under The Curve ◽

High Sensitivity ◽

Ultra Performance Liquid Chromatography ◽

Advanced Neoplasia ◽

Flight Mass Spectrometry

Low invasive tests with high sensitivity for colorectal cancer and advanced precancerous lesions will increase adherence rates, and improve clinical outcomes. We have performed an ultra-performance liquid chromatography/time-of-flight mass spectrometry (UPLC-(TOF) MS)-based metabolomics study to identify faecal biomarkers for the detection of patients with advanced neoplasia. A cohort of 80 patients with advanced neoplasia (40 advanced adenomas and 40 colorectal cancers) and 49 healthy subjects were analysed in the study. We evaluated the faecal levels of 105 metabolites including glycerolipids, glycerophospholipids, sterol lipids and sphingolipids. We found 18 metabolites that were significantly altered in patients with advanced neoplasia compared to controls. The combinations of seven metabolites including ChoE(18:1), ChoE(18:2), ChoE(20:4), PE(16:0/18:1), SM(d18:1/23:0), SM(42:3) and TG(54:1), discriminated advanced neoplasia patients from healthy controls. These seven metabolites were employed to construct a predictive model that provides an area under the curve (AUC) median value of 0.821. The inclusion of faecal haemoglobin concentration in the metabolomics signature improved the predictive model to an AUC of 0.885. In silico gene expression analysis of tumour tissue supports our results and puts the differentially expressed metabolites into biological context, showing that glycerolipids and sphingolipids metabolism and GPI-anchor biosynthesis pathways may play a role in tumour progression.

Download Full-text

Nomogram to predict the progression of patients with primary membranous nephropathy and nephrotic syndrome

International Urology and Nephrology ◽

10.1007/s11255-021-02859-x ◽

2021 ◽

Author(s):

Lili Liu ◽

Haitao Wang ◽

Ban Zhao ◽

Xin Liu ◽

Ying Sun ◽

...

Keyword(s):

Risk Factors ◽

Nephrotic Syndrome ◽

Membranous Nephropathy ◽

Predictive Value ◽

Area Under The Curve ◽

Predictive Ability ◽

Stepwise Logistic Regression ◽

Stepwise Logistic Regression Analysis ◽

Discriminative Ability

Abstract Background The outcome of patients with primary membranous nephropathy (pMN) who present with nephrotic syndrome (NS) is variable and difficult to predict. The goal of this study was to develop a nomogram to predict the risk of progression for specific individuals. Methods This retrospective study involved biopsy-proven patients with pMN and NS treated between January 2012 and June 2018. The primary outcome of our investigation was progression, defined as a reduction of estimated glomerular filtration rate (eGFR) that was equal to or over 20% compared with baseline at the end of follow-up or the onset of end-stage renal disease (ESRD). We used backwards stepwise logistic regression analysis to create a nomogram to predict prognosis. The model was validated internally using bootstrap resampling. Results A total of 111 patients were enrolled. After a median follow-up of 40.0 months (range 12–92 months), 18.9% (21/111) patients showed progression. Backwards stepwise selection using the Akaike information criterion (AIC) identified the following four variables as independent risk factors for progression, which were all used in the nomogram: age ≥ 65 years [odds ratio (OR) 7.004; 95% confidence interval (CI) 1.783–27.505; p = 0.005], Ln (sPLA2R-Ab) (OR 2.150; 95% CI 1.293–3.577; p = 0.003), Ln (proteinuria) (OR 5.939; 95% CI 1.055–33.436; p = 0.043) and Ln (Uα1m/Cr) (OR 2.808; 95% CI 1.035–7.619; p = 0.043). The discriminative ability and calibration of the nomogram revealed good predictive ability, as indicated by a C-index of 0.888 (95% CI 0.814–0.940) and a bootstrap-corrected C-index of 0.869; calibration curves were also well fitted. A receiver operating characteristic (ROC) curve for the nomogram score revealed significantly better discrimination than each of the three risk factors alone, including Ln (sPLA2R-Ab) [area under the curve (AUC) 0.769], Ln (proteinuria) (AUC 0.653) and Ln (Uα1m) (AUC 0.781) in the prediction of progression (p < 0.05). The optimal cutoff value of the nomogram score was 117.8 with a positive predictive value of 44.4% and a negative predictive value of 98.5%. Conclusion The nomogram successfully achieved good predictive ability of progression for patients with pMN who present with NS. It can therefore help clinicians to individualize treatment plans and improve the outcome of pMN.

Download Full-text

Circulating miRNA as Biomarkers for Colorectal Cancer Diagnosis and Liver Metastasis

Diagnostics ◽

10.3390/diagnostics11020341 ◽

2021 ◽

Vol 11 (2) ◽

pp. 341

Author(s):

Farah J. Nassar ◽

Zahraa S. Msheik ◽

Maha M. Itani ◽

Remie El Helou ◽

Ruba Hadla ◽

...

Keyword(s):

Colorectal Cancer ◽

Liver Metastasis ◽

Diagnostic Accuracy ◽

Area Under The Curve ◽

Stage Iv ◽

Screening Tests ◽

Operating Characteristics ◽

Diagnostic Biomarkers ◽

Circulating Mirna ◽

Non Invasive

Colorectal cancer (CRC) is the second leading cause of cancer deaths worldwide. Stage IV CRC patients have poor prognosis with a five-year survival rate of 14%. Liver metastasis is the main cause of mortality in CRC patients. Since current screening tests have several drawbacks, effective stable non-invasive biomarkers such as microRNA (miRNA) are needed. We aim to investigate the expression of miRNA (miR-21, miR-19a, miR-23a, miR-29a, miR-145, miR-203, miR-155, miR-210, miR-31, and miR-345) in the plasma of 62 Lebanese Stage IV CRC patients and 44 healthy subjects using RT-qPCR, as well as to evaluate their potential for diagnosis of advanced CRC and its liver metastasis using the Receiver Operating Characteristics (ROC) curve. miR-21, miR-145, miR-203, miR-155, miR-210, miR-31, and miR-345 were significantly upregulated in the plasma of surgery naïve CRC patients when compared to healthy individuals. We identified two panels of miRNA that could be used for diagnosis of Stage IV CRC (miR-21 and miR-210) with an area under the curve (AUC) of 0.731 and diagnostic accuracy of 69% and liver metastasis (miR-210 and miR-203) with an AUC = 0.833 and diagnostic accuracy of 72%. Panels of specific circulating miRNA, which require further validation, could be potential non-invasive diagnostic biomarkers for CRC and liver metastasis.

Download Full-text

Artificial intelligence applied on chest X-ray can aid in the diagnosis of COVID-19 infection: a first experience from Lombardy, Italy

10.1101/2020.04.08.20040907 ◽

2020 ◽

Cited By ~ 9

Author(s):

Isabella Castiglioni ◽

Davide Ippolito ◽

Matteo Interlenghi ◽

Caterina Beatrice Monti ◽

Christian Salvatore ◽

...

Keyword(s):

Artificial Intelligence ◽

Neural Networks ◽

Convolutional Neural Networks ◽

Area Under The Curve ◽

Training Dataset ◽

List Type ◽

X Rays ◽

Independent Dataset ◽

X Ray ◽

Chest X Ray

AbstractObjectivesWe tested artificial intelligence (AI) to support the diagnosis of COVID-19 using chest X-ray (CXR). Diagnostic performance was computed for a system trained on CXRs of Italian subjects from two hospitals in Lombardy, Italy.MethodsWe used for training and internal testing an ensemble of ten convolutional neural networks (CNNs) with mainly bedside CXRs of 250 COVID-19 and 250 non-COVID-19 subjects from two hospitals. We then tested such system on bedside CXRs of an independent group of 110 patients (74 COVID-19, 36 non-COVID-19) from one of the two hospitals. A retrospective reading was performed by two radiologists in the absence of any clinical information, with the aim to differentiate COVID-19 from non-COVID-19 patients. Real-time polymerase chain reaction served as reference standard.ResultsAt 10-fold cross-validation, our AI model classified COVID-19 and non COVID-19 patients with 0.78 sensitivity (95% confidence interval [CI] 0.74–0.81), 0.82 specificity (95% CI 0.78–0.85) and 0.89 area under the curve (AUC) (95% CI 0.86–0.91). For the independent dataset, AI showed 0.80 sensitivity (95% CI 0.72–0.86) (59/74), 0.81 specificity (29/36) (95% CI 0.73–0.87), and 0.81 AUC (95% CI 0.73– 0.87). Radiologists’ reading obtained 0.63 sensitivity (95% CI 0.52–0.74) and 0.78 specificity (95% CI 0.61–0.90) in one centre and 0.64 sensitivity (95% CI 0.52–0.74) and 0.86 specificity (95% CI 0.71–0.95) in the other.ConclusionsThis preliminary experience based on ten CNNs trained on a limited training dataset shows an interesting potential of AI for COVID-19 diagnosis. Such tool is in training with new CXRs to further increase its performance.Key pointsArtificial intelligence based on convolutional neural networks was preliminary applied to chest-X-rays of patients suspected to be infected by COVID-19.Convolutional neural networks trained on a limited dataset of 250 COVID-19 and 250 non-COVID-19 were tested on an independent dataset of 110 patients suspected for COVID-19 infection and provided a balanced performance with 0.80 sensitivity and 0.81 specificity.Training on larger multi-institutional datasets may allow this tool to increase its performance.

Download Full-text